To what extent do female bureaucrats support female political leaders? How does a scandal where a predominantly male institution investigates a female politician impact bureaucratic support? Research on women's support for female politicians is mixed. Some research suggests that women support female politicians (Dolan Reference Dolan2004). Other research shows no link among the broader population between voter gender and preference for female candidates (Campbell and Heath Reference Campbell and Heath2017; McElroy and Marsh Reference McElroy and Marsh2010). Most research, however, suggests that women's support for women politicians is context-specific (Campbell and Heath Reference Campbell and Heath2017). One particularly important contextual factor is a scandal. Research shows that women may be punished more heavily for scandals than men, particularly for corruption scandals (Barnes, Beaulieu, and Saxton Reference Barnes, Beaulieu and Saxton2020; Eggers, Vivyan, and Wagner Reference Eggers, Vivyan and Wagner2018).
While illuminating, existing studies do not examine how real-world circumstances, such as scandals, drive gender-based solidarity particularly within governments. Understanding the impact of circumstances is important. Past work suggests that different contexts, such as socioeconomic status, culture, and institutional rules, may drive variation in support for female politicians (Paxton and Hughes Reference Paxton and Hughes2014). However, as differences in the success of female politicians across timings within countries suggest, support for women can be highly contingent even within a given set of institutions or socioeconomic environments (Dolan Reference Dolan1998). This suggests that, in any given country, female bureaucrats might be more likely to rally around female politicians on some situations than in others. This article asks when that might be the case.
Drawing on collective identity theory, we suggest that gender solidarity may increase when female bureaucrats are primed to see gender as salient. Collective identity theory suggests that this is more likely to occur when events increase the perceived sense of injustice towards a group, which in turn can enhance the salience of a group's identity and cause the group to band together to oppose the perceived injustice (Duncan Reference Duncan, Deaux and Snyder2018). Such solidarity may also make more women feel a sense of shared experience with the woman in question. Logically, a lack of perceived injustice will fail to activate group solidarity. For this reason, we suggest that contested corruption allegations against a high-profile female politician could activate gender-based identity. Such solidarity is likely to be particularly pronounced when the women are also public servants and thus share cumulating identities (e.g., Eagly and Heilman Reference Eagly and Heilman2016; Lee and Park Reference Lee and Park2021).
To test this, we take advantage of a survey experiment conducted on male and female bureaucrats before and after the impeachment of South Korean female president Park Geun-Hye. Using a list experiment to gauge responsiveness to President Park, we show that prior to her impeachment, male and female bureaucrats were roughly equivalent in their support for President Park. However, post-impeachment, women increased their support for the president, while men did not change in their assessment, thus leading to a sizeable difference in support between the two groups. Mediation analysis provides suggestive evidence that this effect is driven in part by a heightened perception of injustice amongst female bureaucrats. While there are reasons that we enumerate below to be cautious as to whether these findings would generalize to the broader population, these findings have important implications for how we understand gender identity in democratic contexts. In particular, gender-based solidarity is likely to vary not only according to macro-level differences across countries, but also due to event-based changes within governments.
A collective identity explanation for female bureaucrats’ support for female political leaders
Why might women support female political leaders in some contexts but not in others? Scholars adduce a number of explanations to explain women's support for female politicians. Some suggest that gender operates solely through partisanship (Cook Reference Cook, Cook, Thomas and Wilcox1994). Others suggest that such support is driven by a shared group identity (Plutzer and Zipp Reference Plutzer and Zipp1996). Other work theorizes that it is because female politicians promote issues that are of greater importance to women (Paolino Reference Paolino1995) or that they represent an appealing “outsider” image during elections when voters generally are dissatisfied with the status quo (Dolan Reference Dolan1998). However, in general, this literature is inconclusive and generally finds that women's support for women candidates is not uniform and is highly context specific (Dittmar Reference Dittmar2016).
Still other work focuses on why voters, whether men or women, might vote for female candidates. Some find that female politicians tend to be less corrupt than male politicians (Swamy et al. Reference Swamy, Knack, Lee and Azfar2001), particularly when accountability mechanisms are strong (Esarey and Schwindt-Bayer Reference Esarey and Schwindt-Bayer2018). Finally, some research suggests that sexism can drive increased voter punishment for female politicians compared to male politicians participating in a scandal (Barnes, Beaulieu, and Saxton Reference Barnes, Beaulieu and Saxton2020). This relates to a larger literature showing that voters tend to perceive women as less corrupt than men due to their outsider image and perception that women are more risk-averse (Barnes, Beaulieu, and Saxton Reference Barnes, Beaulieu and Saxton2017).
We do not disagree with this work, and indeed, all these forces likely operate. However, we also suggest that this explanation may be incomplete and fail to account for shifts in female bureaucrats’ support for female political leaders based on real world events. One study close to our own suggests that women may in fact punish female politicians for engaging in corruption (Eggers, Vivyan, and Wagner Reference Eggers, Vivyan and Wagner2018). Due to its experimental setting, their study has an important advantage: it can directly compare perceptions of corruption investigations leveled against male and female politicians. Unfortunately, our study, using a real-world corruption accusation, cannot directly make this comparison. At the same time, the experimental setting in Eggers, Vivyan, and Wagner (Reference Eggers, Vivyan and Wagner2018) faces an important drawback: respondents were told the candidate was corrupt, hence there was no ambiguity about the veracity of the allegation. Real world situations differ, because observers must decide whether the allegation and process of handling the case is fair. In short, not all respondents may process a given scandal equally. Thus, a non-hypothetical setting on different subsets of female respondents may lead to different outcomes.
With this in mind, we suggest that contingent factors may raise the salience of gender identity in a way that could increase solidarity and support for female politicians in the short-term. Psychologists find that when groups perceive an injustice towards their group or when latent injustices are exposed, this can increase group consciousness and willingness to engage in collective action on behalf of the group (Duncan Reference Duncan, Deaux and Snyder2018). That is, a female bureaucrat's perception that female chief executives face harsher punishment because of their gender may lead to solidarity with the leaders of the same gender. Moreover, scholars of executive politics and public administration find that bureaucrats tend to act on their strong belief and perception of political events and elites. Building on evidence from various disciplines, we suggest, holding all other factors constant, a salient event occurring to a female president perceived as unfair based on gender should drive greater solidarity. This leads to our fundamental hypothesis that political punishment of female leaders should lead to greater support from female bureaucrats relative to male counterparts within government. Male bureaucrats, on the other hand, will respond to the political event by reducing their support due to concerns about the allegations.
Research design and context
We use the timing of the impeachment in 2016 of President Park Geun-Hye, South Korea's first female president, to assess how the salience of gender impacts support for the president differentially across women and men. We use a list experiment of more than 1,000 civil servants in South Korea to test the effect. Our sample of bureaucrats instead of voters has important implications for how we generalize our results. While we might like to examine the impact of the impeachment on voters in general, we are taking advantage of a survey that happened to include questions that could measure support for President Park and spanned the period when the impeachment proceedings unfolded. While different from typical studies, this sample offers the unique advantage of assessing the impact of scandal on the understudied bureaucratic population, which is arguably a more important population given their impact on policy implementation. With that said, the bureaucratic sample may generate stronger findings than if we conducted the sample on voters. Given that bureaucrats have greater political awareness than the more general population and that women bureaucrats in particular may feel more solidarity to a fellow civil servant in President Park than women not involved in government, the solidarity expressed by the women in our sample may be greater than would be the case for the general population. Therefore, we should be cautious in suggesting that our findings would necessarily translate to the broader population. At the same time, the sample is useful for examining the rallying effect, given that we expect bureaucrats to have greater knowledge of the role of the president in policy formation.
Regarding the context, for the survey to identify the impact of the impeachment, it is important that the outcome of the legislative vote for an impeachment proposal not be known ahead of time. If this is the case, respondents would have already perceived the injustice ahead of the vote thus leading to no effect. On this point, existing work on Park's impeachment shows that impeachment was not an obvious option to lawmakers and the political state of play was “highly fluid and subject to change,” as there were at least “three possible scenarios, each with variants.”Footnote 1 Most notably, the institutional context led to a political stalemate in the legislature: the passage of the impeachment proposal needed consent by two-thirds of the legislature, but the president's party, which had more than 40 percent of legislative seats, was against the proposal. On December 9, 2016, when there was a floor vote on the impeachment proposal, a minority faction of the president's party defected to vote for Park's impeachment, which led to two-thirds of the National Assembly consenting to impeach Park.
In terms of the salience of Park's impeachment for women inside the civil service, her impeachment was seen as having important ramifications for gender equality in South Korea. The country is a patriarchal society for its level of development, which made Park's initial election significant for gender equality advocates. Despite her mixed record on advocating for women, some evidence shows women supported her in greater numbers in 2012 than they would for a typical conservative party candidate, suggesting that gender played a role in her election (Lee and Jalalzai Reference Lee and Jalalzai2017). Additionally, many advocates expressed concern that impeachment by an overwhelmingly male legislature would set South Korea's women's equality movement back.Footnote 2 Furthermore, others suggest gender played a role in her impeachment, asserting that male presidents would not have been impeached for similar wrongdoing.Footnote 3 Such commentaries were common, indicating that gender was a prominent issue regarding the impeachment.
Turning to the survey instrument, to construct our dependent variable, we use a list experiment question to assess whether or not bureaucratic respondents find the president's preferences important when making decisions or policies. We use a list experiment because it could be potentially sensitive to ask questions about support for a president when that president is in power.Footnote 4 A list experiment reduces sensitivity by randomly providing respondents with two lists of activities—one which includes only non-sensitive items and one which includes the sensitive item in addition to the non-sensitive items. The control group receives the non-sensitive list, and the treatment group receives the non-sensitive list with the additional sensitive item. Respondents are asked to provide the number of items on the list that apply to them. The researcher can then estimate the prevalence of the sensitive item by comparting the number of items that apply to the control and treatment (Blair and Imai Reference Blair and Imai2012). Because list experiments require greater power, our results are actually stronger when we use a direct question as the outcome variable.Footnote 5
Our survey took place in South Korea between November 23 and December 30, 2016, the periods before and after the National Assembly impeached President Park on December 9, 2016.Footnote 6 As part of a survey initially designed for other purposes,Footnote 7 by chance about half of the sample was presented with the experiment before an impeachment proposal was levied against Park; the other half was presented with the same experiment after the impeachment was ratified.Footnote 8 Respondents were randomly assigned to either control or treatment groups within these pre- and post-impeachment samples. In our study, we asked the following question to the control group:
Below several elements are listed. How many of these elements do you consider important when making and implementing government policy decisions related to people's needs today? You do not need to say which ones are important, only say HOW MANY you agree with.
(1) Major public opinion reflected through media
(2) Other central ministries' positions
(3) Influence of civic and interest groups
The question asks the treatment group the same question as the control group, with the exception that the treatment group received an additional item: “The president's preferences presented from the Blue House.” Due to the high sensitivity of listing an item that asks civil servants to assess President Park, our treatment item is rather implicit. However, since our survey question explicitly asks about civil servants’ consideration in making and implementing policy decisions today, it should be clear that civil servants had President Park in mind when they chose this item or not.
The second treatment group received the item: “Views reflecting a majority of the National Assembly.” While initially designed to assess responsiveness to the legislature, this second group can be seen as a placebo to ensure that the difference in male versus female bureaucrats’ support is not due to more general changes in attitudes towards political accountability. By comparing the proportion of respondents selecting the treatment item before and after a female president's impeachment, we provide information about changes in bureaucrats’ responsiveness to the president across civil servant gender.
An important question is whether the instrument proxies for support for Park. While many civil servants view their job of representing the preferences of the president as a general duty, we suggest that this should militate against the difference in support for the first treatment statement pre- and post-treatment. However, we argue that at least some of the support for the statement is due to the underlying degree to which the civil servant supports the current occupant of the Blue House.Footnote 9 Indeed, as our survey shows, co-partisans did also support Park more after impeachment than members of the opposition party.Footnote 10
Results
In this section, we compare the number of items between the treatment and control groups from the list experiment for male and female civil servants pre- and post-impeachment. We use maximum likelihood estimates to formally test our argument (see Table A1 in the appendix for a summary of the data). The distribution of response values is normal across the two groups before and after impeachment.
Based on the observed data, we report the main univariate results from our experiment in Table 1. The results suggest that impeachment impacts support differentially across civil servant gender. While male support declined by 20 percentage points, female bureaucrats increased support by more than 40 percentage points. The detailed interpretation of the univariate results is provided in Appendix Section C.
Table 1 Mean response to control and treatment items and its difference before and after impeachment, by gender
***p < .01; **p < .05; *p < .10. Standard errors in parentheses.
In order to further verify the results of our univariate analysis, we run a multivariate regression analysis based on the maximum likelihood estimator (MLE) adopted by Blair and Imai (Reference Blair and Imai2012). This additional analysis is useful in the sense that we can estimate the population proportion of respondents of different genders, who are likely to choose the sensitive items, accounting for other individual characteristics.
Before conducting a multivariate statistical analysis, we first balance pre- and post-impeachment samples on a variety of characteristics using coarsened exact matching (CEM) and we pair civil servants surveyed before and after impeachment so that they match on a variety of covariates including demographic (education, age, gender), occupational (private sector experience, job category, recruitment path), and agency (agency head gender) characteristics, as well as political views. To employ CEM, matches are selected by first coarsening the individual characteristics of civil servants into discrete categories and then pairing pre- and post-motion samples so that they are exact matches on all included covariates. CEM performs better than other matching approaches, such as propensity score matching, in its ability to reduce bias and estimation errors (Iacus, King, and Porro Reference Iacus, King and Porro2012).
By comparing pre- and post-motion samples that otherwise share similar characteristics, we can better isolate the direct effect of bureaucrat gender on their support for the president, while controlling for other individual differences that might be highly associated with gender, such as age and past work experience, and also influence the likelihood of support. With a balanced subset of data, we then estimate the effect of gender using the MLE. Appendix Table A3 provides descriptive statistics for our independent and control variables, including information on how each variable is coarsened. F-test results (see Table A3) show no differences among covariates pre- and post-impeachment groups at conventional levels of statistical significance.
In Table 2, we examine the effect of civil servant gender on bureaucratic responsiveness in our matching design. An exactly balanced subset of data allows us to estimate the effect of gender using MLE without the necessity of controlling for other covariates described above. Yet, we also report the results of full model specifications with other control variables in Appendix Table A6. Presented are the coefficients of the maximum likelihood estimator for three model specifications (for two treatment groups and one control group) where the dependent variables are the likelihood of supporting the president (Model 1) and the legislative majority (Model 2) pre- and post-impeachment; and the independent variables are the interaction terms between gender (1 = female, 0 = male) and impeachment (1 = post-impeachment, 0 = pre-impeachment). We demonstrate the results of the difference in estimated proportions of male and female respondents affirmatively in Figure 1.
Figure 1 Difference in estimated proportions of male and female bureaucrats answering the “president's treatment” item affirmatively and their difference
Note: Results based on regression models from our exact matching design in Table 2. Positive and negative estimates show male and female bureaucrats, respectively, have more positive perceptions toward the treatment.
Table 2 Multivariate regression analysis: Maximum likelihood estimator
Note: Estimated coefficients are based on a maximum likelihood estimator where the dependent variables are whether or not “the president's preferences” are considered important to respondents in policy decision making and implementation (Model 1) and whether or not “opinions reflecting a majority of the National Assembly” are considered important to respondents in policy decision making and implementation (Model 2).
We continue to find the differential effect of civil servant gender on bureaucratic responsiveness after impeachment: Female bureaucrats increased support for the president, whereas male bureaucrats’ support was largely unchanged. Before impeachment, male bureaucrats were slightly more supportive of the president than female civil servants, but not a conventional level of significance. However, after impeachment, female bureaucrats were more supportive of the president, with an estimated difference of 38.2 percentage points (78.4 percent for females vs. 40.2 percent for males), which is significant at the 95 percent confidence level.
Regarding bureaucratic responsiveness to the legislative majority based on our second treatment, neither male nor female bureaucrats become more or less accountable to the legislative majority, as estimated differences in the multivariate results are not significant at the 95 percent confidence level in pre- and post-motion periods. These results concerning our second treatment are shown in Figure 2. When the gender difference in responses to our treatment items after impeachment is compared, the estimated difference is -0.382 (with the 95 percent confidence interval of [-0.752, -0.013]) for the president treatment, whereas the estimated difference is -0.299 (with the 95 percent confidence interval of [-0.827, 0.228]) for the legislative majority placebo. This indicates that presidential treatment has a significant differential effect based on gender while the legislative treatment does not. With that said, a comparison of the results between the placebo and the treatment does also indicate that there is no significant difference between the effect of gender in the presidential treatment and the legislative placebo. While that could generate some caution that our survey is picking up something specific to the presidential treatment, we should also note that both the treatment and placebo are using list experiments, require a great deal of power to detect a difference between the two results (Blair, Coppock, and Moor Reference Blair, Coppock and Moor2020). Thus, with a direct question, it is possible that the differential effect between placebo and treatment would be greater.
Figure 2 Difference in estimated proportions of male and female bureaucrats answering the “legislative majority's treatment” item affirmatively and their difference
Note: Results based on regression models from our exact matching design in Table 2. Positive and negative estimates show male and female bureaucrats, respectively, have more positive perceptions toward the treatment.
To assess the causal mechanism, we theorize that Park's impeachment led female bureaucrats to perceive more injustice in the political system, which caused them to support the president. Therefore, we further analyze whether bureaucrats’ different responses across gender were mediated by their perception of injustice towards her impeachment. For this analysis, we adopt a general approach to casual mediation analysis (Imai et al. Reference Imai, Keele and Tingley2010), with the mediating variable relying on a question asking respondents to assess how much they agreed or disagreed with the following statement: “I don't consider current major political phenomenon to be fair.” As this question was not explicitly designed for our study, it may not perfectly capture a sense of injustice towards women in politics. Ideally, our question would ask more explicitly whether women are treated fairly in politics. However, we believe at least some of the respondents will have impeachment in mind when answering this question. For the dependent variable, we use the direct question to generate a binary variable about support for the president.
With this variable, we run a causal mediation analysis to show 1) whether her impeachment led to any change in bureaucrats’ perception of injustice and 2) how any change in the perception of injustice affected bureaucrats’ support for a female president. In Table 3, we show the results of our causal mediation analysis for female and male bureaucrats. The results in Table 3 suggest that impeachment increased female bureaucrats’ perception of injustice (Model 1) but not for men (Model 3), thus confirming the first link in the causal mediation analysis. Impeachment generated a sense of injustice for women but not men.Footnote 11 This in turn caused women to increase their support for a female president (Model 2), with the mediating variable explaining an estimated 7 percent of the effect of impeachment on increased support for Park. In contrast, impeachment did not change male bureaucrats’ perception of injustice (Model 3) or their support for a female president (Model 4). In Appendix Section F, we discuss these results in more detail.
Table 3 Causal mediation analysis
***p < .01; **p < .05; *p < .10.
Note: In Models 1 and 3, we employ OLS models where the dependent variables are to what extent respondents consider the current major political phenomenon unfair. In Models 2 and 4, we employ logit models where the dependent variables are whether the president's preferences are considered important to respondents in policy decision making and implementation.
Conclusion
Building on collective identity theory, our theory suggests that female bureaucrats may rally around a female politician accused of corruption if that allegation is seen as unfair. In our study, after President Park's impeachment, female bureaucrats display higher support for the president, whereas men did not change their support. Our findings suggest that salient punishment of high-profile female political leaders can drive support from female bureaucrats who see those politicians as part of their identity group. In short, contingent events can drive increased female bureaucrats’ support for female politicians, even over a short period of time. While the collective identity effect may be more marked among women inside the government, our data still offer a unique opportunity to test the plausibility of the theory.
Our findings leave open the possibility that bureaucrats may be more aware of misconduct by a female president or perceive similar behavior by male and female political leaders differently, which should be addressed in future research on civil servants’ behavior in the context of a male president's impeachment. Furthermore, our outcome variable, which takes advantage of a survey experiment asking about support for a president's policies rather than support for President Park directly, could be excessively indirect. Future work could include a more direct measure. With that said, we believe this indicator should be substantially driven by the support for the occupant of the Blue House. As such, we believe our study provides the first experimental evidence that male and female bureaucrats differ in maintaining their responsiveness to a president in a context where bureaucrats’ perception of injustice towards their political leader can vary.
Funding
This work was supported by Ministry of Education of Republic of Korea and National Research Foundation of Korea (BK21FOUR Toward Empathic Innovation: Through Platform Governance Education & Research Programs: #4199990114294) and Public Performance Management Research Center at Seoul National University (0678-20200040).
Conflict of Interest
The authors declare none.
Supplementary Material
The supplementary material for this article can be found at https://doi.org/10.1017/jea.2022.36
Open access
